On Friday 30 November 2007 01:30, Weddington, Eric wrote: [...] > > Yes nog using -lm indeed adds the 256 bytes > > Oh, how interesting! So there may be a workaround? :-) > > Can you add that bit of information to the bug report? > > We usually tell users that if they are working with any floating point > then they should always use -lm, otherwise all bets are off. > > Also, is there any way that you can confirm that this also works with > the CVS version of avr-libc? Or the 1.5.x version of avr-libc? Both > contain the new FP library.
Hi. Has checked up with version 4.2.2. Addition of the '-lm' option reduces the size up to 770 bytes and eliminates an array in the RAM. It is right: Avr-libc's conversion functions are used, not from libgcc. It is no matter, CVS or 1.4 branch. Float library from Avr-libc 1.5 (or CVS) is very similar to libgcc's one. Such thinks as subnormal numbers, negative zero, infinities and NaNs are supported. So it is possible to say: if you are use 1.5 branch, you can use the '-lm' options always. But if you are use older, 1.4 branch, you can obtain the different results with and without '-lm' option, as the older fplib is not correct for unnormal numbers. Dmitry. _______________________________________________ AVR-libc-dev mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/avr-libc-dev
